基于图注意力网络的购物篮推荐系统  

A Basket Recommendation System Based on Graph Attention Network

作　　者：韩敬争 郭峰[1] Han Jingzheng;Guo Feng(Shandong Provincial Key Laboratory of Optical Communication Science and Technology,Liaocheng University,Liaocheng 252000)

机构地区：[1]聊城大学山东省光通信科学与技术重点实验室,聊城252000

出　　处：《现代计算机》2022年第22期86-92,共7页Modern Computer

基　　金：国家自然科学基金青年科学基金项目(11504153):含能材料冲击Hugoniot曲线的分子动力学方法理论研究。

摘　　要：目前已经有很多关于推荐系统的工作取得了很好的成绩。然而大多数工作都集中在对购物篮中单个项目进行排序,关于下一个购物篮的推荐方案却很少。GATBR是一个基于图注意力网络的购物篮推荐模型,它分别对购物篮和其中项目建模。网络中的注意力模块针对给定项目的历史行为,学习其中复杂的交互关系,最后将其合并表示为超图,为目标用户提供最有可能出现在下一个购物篮中的项目参考。通过在Yoochoose和TaFeng数据集上进行实验,GATBR表现出高于其它现有方案的预测精度。At present,there have been a lot of work on the recommendation system and achieved good results.However,most of the work focuses on sorting individual items in the basket,and there are few recommended schemes for the next basket.GATBR is a basket recommendation model based on graph attention network(GAT),which models basket and its items respectively.The attention module in the network learns the complex interaction relationship of a given item according to its historical behavior.Finally,it is combined and expressed as a hypergraph to provide the target user with the item reference that is most likely to appear in the next basket.Through experiments on Yoochoose and Tafeng datasets,GATBR shows higher prediction accuracy than other existing schemes.

关 键 词：购物篮推荐 图注意力网络 超图 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]